Learning device, learning method, and learning program

Abstract

A processor is configured to: acquire training data that consists of a learning expression medium and a correct answer label for at least one of a plurality of types of classes included in the learning expression medium; input the learning expression medium to a neural network such that probabilities that each class included in the learning expression medium will be each of the plurality of types of classes are output; integrate the probabilities that each class will be each of the plurality of types of classes on the basis of classes classified by the correct answer label of the training data; and train the neural network on the basis of a loss derived from the integrated probability and the correct answer label of the training data.

What is claimed is: 
     
         1 . A learning device for performing machine learning on a neural network that classifies an expression medium into three or more types of classes, the learning device comprising:
 at least one processor,   wherein the processor is configured to:   acquire training data that consists of a learning expression medium and a correct answer label for at least one of a plurality of types of classes included in the learning expression medium, wherein the learning expression medium is an image and the plurality of types of classes are a plurality of regions including a background in the image;   input the learning expression medium to the neural network such that probabilities that each class included in the learning expression medium will be each of the plurality of types of classes are output;   integrate the probabilities that each class will be each of the plurality of types of classes on the basis of classes classified by the correct answer label of the training data;   train the neural network on the basis of a loss derived from the integrated probability and the correct answer label of the training data; and   add probabilities of classes other than the classes classified by the correct answer label for the learning expression medium and a probability of the background among the probabilities that the classes will be the plurality of types of classes to integrate the probabilities that each class will be each of the plurality of types of classes.   
     
     
         2 . The learning device according to  claim 1 ,
 wherein the classes classified by the correct answer label include two or more of the plurality of types of classes, and   the processor is configured to add probabilities of the two or more classes classified by the correct answer label among the probabilities that the classes will be the plurality of types of classes to integrate the probabilities that each class will be each of the plurality of types of classes.   
     
     
         3 . The learning device according to  claim 1 ,
 wherein the processor is configured to train the neural network using a plurality of training data items having different correct answer labels.   
     
     
         4 . A learning method for performing machine learning on a neural network that classifies an expression medium into three or more types of classes, the learning method comprising:
 acquiring training data that consists of a learning expression medium and a correct answer label for at least one of a plurality of types of classes included in the learning expression medium, wherein the learning expression medium is an image and the plurality of types of classes are a plurality of regions including a background in the image;   inputting the learning expression medium to the neural network such that probabilities that each class included in the learning expression medium will be each of the plurality of types of classes are output;   integrating the probabilities that each class will be each of the plurality of types of classes on the basis of classes classified by the correct answer label of the training data;   training the neural network on the basis of a loss derived from the integrated probability and the correct answer label of the training data; and   adding probabilities of classes other than the classes classified by the correct answer label for the learning expression medium and a probability of the background among the probabilities that the classes will be the plurality of types of classes to integrate the probabilities that each class will be each of the plurality of types of classes.   
     
     
         5 . A non-transitory computer-readable storage medium that stores a learning program causing a computer to execute a learning method for performing machine learning on a neural network that classifies an expression medium into three or more types of classes, the learning program causing the computer to execute:
 a procedure of acquiring training data that consists of a learning expression medium and a correct answer label for at least one of a plurality of types of classes included in the learning expression medium, wherein the learning expression medium is an image and the plurality of types of classes are a plurality of regions including a background in the image;   a procedure of inputting the learning expression medium to the neural network such that probabilities that each class included in the learning expression medium will be each of the plurality of types of classes are output;   a procedure of integrating the probabilities that each class will be each of the plurality of types of classes on the basis of classes classified by the correct answer label of the training data;   a procedure of training the neural network on the basis of a loss derived from the integrated probability and the correct answer label of the training data; and   a procedure of adding probabilities of classes other than the classes classified by the correct answer label for the learning expression medium and a probability of the background among the probabilities that the classes will be the plurality of types of classes to integrate the probabilities that each class will be each of the plurality of types of classes.